SLE patients were diagnosed according to the American College of Rheumatology classification criteria and/or the Systemic Lupus International Collaborating Clinics (SLICC) criteria  (21, 22), and were recruited from the Division of Immunology and Allergy at Centre Hospitalier Universitaire Vaudois (CHUV). Current or past use of belimumab or rituximab was an exclusion criterion. All patients and controls were included in the Swiss Systemic Lupus Erythematosus Cohort Study (SSCS) (23). Disease activity score was measured using the SLE Disease Activity Index (SLEDAI) scoring system. We categorized patients into three groups of disease activity: inactive (SLEDAI 0-3), moderate (SLEDAI 4-10) and active (SLEDAI >10).

Two distinct cohorts were examined: cohort 1 included 28 SLE patients and age-, sex-, and ethnicity-matched healthy controls ( Supplementary Table 1A ). Cohort 2 included 10 patients with SLE, 10 age-, sex-, and ethnicity-matched healthy controls, 10 patients with biopsy-proven sarcoidosis (SAR), 10 patients with Sjögren’s syndrome (SJS; based on the 2002 American-European Classification Criteria) and 10 patients with multiple sclerosis (MS; based on the 2017 McDonald Criteria) ( Supplementary Table 1B ). For MS patients, treatment with corticosteroids within three months before the blood draw was an exclusion criterion.

Analysis of absolute cell count was performed on fresh blood by flow cytometry according to standard diagnostic measurements.

For mass cytometry analysis, peripheral blood mononuclear cells (PBMC) were isolated by density gradient centrifugation (FICOLL 400, Merck, Switzerland), from 22.5 ml peripheral blood, and then cryopreserved in liquid nitrogen.

Samples were stained according to a previously published approach (16). Briefly, cryopreserved PBMC from SLE patients and controls (healthy and autoimmune) were thawed, resuspended in RPMI (completed with 20% heat-inactivated serum). Cells (1 Mio per individual on average) were stained for live/dead with cisplatin 50 µg (5min, room temperature (RT)), barcoded with CD45-metal conjugated antibodies (20min, RT, Supplementary Tables 2A, B ) and then pooled. For cohort 1, two HC and two SLE were pooled, for cohort 2 one HC, SLE, SAR, SJS, MS sample were pooled in each experiment. Next, cells were incubated with metal conjugated antibody mix for the extracellular staining (20min, RT). The panel consisted of 39 markers, including markers for SLAMF receptors and for the main PBMC populations (CD4+ T cells, CD8+ T cells, double negative T cells (DN), B cells, natural killer (NK) cells, dendritic cells (DC) and monocytes) and differentiated subsets ( Supplementary Table 2C ). Cells were washed and fixed with 2.4% paraformaldehyde (10 min, RT). Labeled samples were acquired on a Helios Cytof System (Fluidigm). For each experiment at least 500’000 cells were acquired per patient. Flow cytometry standard (FCS) files were normalized to EQ Four Element calibration beads using CyTOF software.

Data were debarcoded on Cytobank software (Beckman Coulter) and fcs files were generated. The fcs files were then analyzed using FlowJo™ software (version 10.2, Becton, Dickinson and Company). All major PBMC populations and subpopulations were gated according to the gating strategy presented in Supplementary Figure 1 . The data were then processed using GraphPad prism (version 8), R software (version 3.5.1) or Python (version 3.8.5). Statistical analysis was performed with GraphPad prism. Specifications of test exploited and sample size are specified in the figure descriptions. In general, data (cell subset frequencies) were transformed into log₁₀(x+1) and normality was assessed with Shapiro-Wilk test. Two groups were compared using Welch’s T test (or Mann-Whitney T test if normality test failed). One-way ANOVA was used for multiple group comparison with normal distribution and p-values were adjusted for multiple testing using Tukey’s test (comparison between all groups) or Dunnett’s test (comparison to a control group). Correlations were assessed using Pearson’s correlation. All data are presented as mean “±” standard error of the mean (SEM). A p-value ≤ 0.05 was considered statistically significant.

Manually gated cell sub-/populations were imported in R studio environment and processed as previously described (16). Briefly, single cell expression was transformed using hyperbolic inverse sine (with cofactor 5) (24). Dimensionality reduction and 2-dimensional visualization were done using the Barnes-Hut implementation of t-stochastic neighboring embedding algorithm (Rtsne package). Unsupervised clustering analysis was performed on previously gated PBMC using self-organizing map in combination with consensus clustering (FlowSOM package). The parameters used for clustering were SLAMF1, SLAMF3, SLAMF4, SLAMF5, SLAMF6 and SLAMF7. The analysis was repeated on subpopulations of cells to ensure consistency of findings. Manual gating was then performed to confirm the existence of an identified cluster. A minimum of 100 cells was required for a cell subset to be considered for further analysis.

Python (Scikit-Learn library) was used to normalize cell frequencies (min-max normalization) of SLEB1, SLEcTFH and SLESMB population. The normalized frequencies were then summed and averaged to obtain combination of the different measurements. ROC curve analysis was used to determine the ability of these measures to distinguish a patient with SLE from healthy or autoimmune controls. The area under the curve (AUC) represents the accuracy of a measurement in distinguishing SLE from controls, and was therefore used as an indicator of separation between groups. Youden index was used to determine the optimal cut-off to separate SLE patients from controls. This cut-off was then applied to cohort 2 to determine the specificity of the approach in identifying SLE patients among patients affected by other autoimmune diseases.

Informed written consent was obtained from all participants prior to inclusion and the study was approved by the Institutional Review Board (SwissEthics 2017-01434 and 2018-01622), in compliance with the Declaration of Helsinki.

The pathophysiology of systemic lupus erythematosus (SLE) is characterized by alterations of the innate and adaptive immune system. To identify an immune signature for SLE, we performed single cell mass cytometry analysis. We included markers for all major PBMC populations, markers of differentiation and markers of activation. First, we assessed the distribution of the main populations of PBMC in healthy controls (HC) and SLE patients: CD4+ T cells, CD8+ T cells, double negative (DN) T cells, B cells, natural killer (NK) cells, dendritic cells (DC) and monocytes ( Figure 1A and Supplementary Figure 1 ). Consistent with previous studies (25, 26), we observed significant lymphopenia in SLE patients compared with HC ( Figure 1B ) and significant decrease in all lymphocyte subpopulations (including CD4+ T cells, CD8+ T cells, DN T cells, B cells, NK cells and DC) ( Figures 1C, D ), validating our technical approach. No difference was observed in abundance of monocytes in SLE patients compared to HC ( Figure 1C ). Interestingly, there was no association between lymphocyte count and disease severity or treatments ( Supplementary Figure 2 ). We proceeded by analyzing the subpopulations of CD4+ T, CD8+ T and B cells ( Supplementary Figure 1 ). The following populations were considered for CD4+ T cells: CD4+ T naïve, effector memory (EM), central memory (CM), terminally differentiated effector memory (TDEM), Th1, Th2, Th17, circulating T follicular helper cells (cTfh), regulatory T cells (Treg). T helper subsets were defined on the basis of cell surface chemokine receptor expression ( Supplementary Figure 1 ). For CD8+ T cells, naïve, EM, CM and TDEM cells were included. Finally, for B cells, naïve, switch memory (SM), non-switch memory (NSM), double negative (DN) and circulating plasma cells (cPC) were included in the analysis ( Supplementary Figure 1 ). We observed that the frequency of cTfh cells (CD45RO+CXCR5+) and cPC (CD27+CD38+) were significantly increased in patients with SLE ( Figure 1E ). The frequency of NSM B cells (CD27+IgD+) was significantly reduced in SLE patients. No other significant alterations in subset frequency were observed.

Several studies indicate that SLAMF receptors play a role in the pathophysiology of SLE, as mentioned above (4). We hypothesize that SLAMF receptors expression defines an immune signature unique to SLE. To investigate this, we first examined the individual expression of each SLAMF receptor (SLAMF1, SLAMF3, SLAMF4, SLAMF5, SLAMF6, SLAMF7) on all main populations and subpopulation of PBMC from SLE patients included in cohort 1 ( Figure 2A ).

SLE patients showed a significant increase in the frequency of CD4+ T cells- and B cells-expressing SLAMF1, as well as CD4+ T cells-, B cells- and monocytes-expressing SLAMF7. Furthermore, there was a decrease in the frequency of DN T cells positive for SLAMF3 and SLAMF4, and of the percentage of DN T cells-, B cells- and monocytes-expressing SLAMF6 ( Figure 2B ). Next, we investigated SLAMF receptors expression on CD4+ T cell, CD8+ T cell and B cell subsets. We found that the percentage of SLAMF1-expressing cells was increased in all SLE CD4+ T cell subsets, including naïve T cells, CM, EM, TDEM, Th1, Th2, Th17, Treg and cTfh cells. Furthermore, the frequency of CD4+ TDEM-expressing SLAMF7 was significantly increased in SLE ( Figure 2C ). No significant alteration was observed in the expression of SLAMF receptors in SLE CD8+ T cell subsets ( Supplementary Figure 3 ). Analysis of SLE B cell subsets indicated an increase in the frequency of naïve, NSM, SM and DN (CD27- IgD-) B cells-expressing SLAMF1. In addition, the frequencies of SM and DN B cells-expressing SLAMF7 were increased in SLE, whereas naïve B cells-expressing SLAMF6 were reduced in SLE patients compared to HC ( Figure 2C ).

Given the suggested relationship between SLAMF receptor expression and the pathophysiology of SLE, we questioned whether single SLAMF expression could serve as marker for disease activity. To answer this question, we first evaluated the individual expression of SLAMF on each population ( Supplementary Figure 4 ) and subpopulation of PBMC. This analysis showed that the frequency of NK cells expressing SLAMF4 NK cells ( Figure 3A ) and of monocytes expressing SLAMF4 ( Figure 3B ) inversely correlated with disease activity. Furthermore, the percentage of TDEM CD4+ T expressing SLAMF1 positively correlated with SLE disease activity ( Figure 3C ).

We evaluated the simultaneous expression of all SLAMF receptors at single-cell level ( Figure 4A ). To run this analysis, we performed unsupervised clustering analysis based on SLAMF expression on pre-gated major PBMC populations. This analysis was followed by unbiased clustering analysis of pre-gated subpopulations of CD4+ T cells, CD8+ T cells, B cells, NK cells, monocytes, and dendritic cells. Populations that were consistently discovered after applying sequential unbiased analysis were manually gated. Based on their relative cell abundance they may be of biological significance ( Supplementary Figure 6A ).

Accordingly, our analysis of CD4+ T identified the following cell subsets, which did not differ in their frequency between SLE and HC, as potentially relevant: naïve CD4+T cells and CM CD4+ T cells co-expressing SLAMF3 and SLAMF6, Th1 CD4+ T cells co-expressing SLAMF1, SLAMF3 and SLAMF6, Th2 CD4+ T cells co-expressing SLAMF5 and SLAMF6, Th17 CD4+ T cells co-expressing SLAMF3, SLAMF5 and SLAMF6 ( Supplementary Figure 5 ). The presence of the following populations was confirmed after manual gating and their frequency was increased in SLE patients compared to HC: EM CD4+ T cells co-expressing SLAMF1, SLAMF3 and SLAMF5, Treg (CD127-CD25high) CD4+ T cells co-expressing SLAMF1 and SLAMF5, and cTFH CD4+ T cells co-expressing SLAMF1, SLAMF3, SLAMF5 and SLAMF6 ( Figure 4B ). Finally, Th1 CD4+ T cells co-expressing SLAMF3 and SLAMF6 were significantly decreased in SLE patients ( Figure 4B ). The frequency of these popualtions was not directly impacted by disease activity or treatments ( Supplementary Figure 7 )

The analysis of CD8+ T cells and subsets identified that EM CD8+ T cell co-expressing SLAMF1, SLAMF3, SLAMF6 and SLAMF7, were significantly decreased in SLE patients ( Figure 4B ). The frequency of these popualtions was not directly impacted by disease activity or treatments ( Supplementary Figure 7 )

Analysis of B cells and B cell subsets, identified one cell subset as potentially relevant, whose frequency was not significantly different between HC and SLE patients: naïve B cells co-expressing SLAMF3 and SLAMF6 ( Supplementary Figure 5 ). Furthermore, naïve B cells co-expressing SLAMF1 and SLAMF3 and SM B cells co-expressing SLAMF1, SLAMF3, SLAMF5 and SLAMF6 were significantly increased in SLE, while cPC co-expressing SLAMF4 and SLAMF6 were reduced ( Figure 4B ). The frequency of these popualtions was not directly impacted by disease activity or treatments ( Supplementary Figure 7 )

From our analysis, we did not identify any SLAMF-based clusters in NK cells, DC and monocytes that exhibit altered frequency in patients with SLE compared to HC. However, CD16+PD1+ monocytes co-expressing SLAMF1, SLAMF5 and SLAMF7+ showed a tendency to be increased in patient with SLE. Furthermore, CD16high NK cells co-expressing SLAMF4, SLAMF6 and SLAMF7 are consistently identified by unbiased clustering analysis and their presence was confirmed by manual gating ( Supplementary Figure 5 ).

In order to identify an immune signature specific to SLE, we considered a second cohort of patients (cohort 2), which included patients with SLE, HC and patients with the following autoimmune diseases: sarcoidosis (SAR), Sjögren’s syndrome (SJS) and multiple sclerosis (MS) ( Figure 5A ).

In patients included in the cohort 2, we examined SLAMF-based cell populations that were identified in the cohort 1. We first focused our analysis on single SLAMF receptor expression. We observed that, among all the populations of interest identified in the cohort 1, only B cells-expressing SLAMF1 (identified as SLEB1) were significantly increased in SLE compared to healthy and autoimmune diseases controls ( Figure 5B ). Then, we examined the frequencies of population defined by the co-expression of multiple SLAMF receptors as characteristics of SLE in cohort 1. This analysis showed that two populations are significantly increased in SLE patients compared to healthy and autoimmune diseases controls ( Figure 5C ): SM B cells co-expressing SLAMF1, SLAMF3, SLAMF5 and SLAMF6 (identified as SLESMB) and cTfh CD4+ T cells co-expressing SLAMF1, SLAMF3, SLAMF5 and SLAMF6 receptors (identified as SLEcTFH).

Overall, analysis of single-expression and co-expression of SLAMF receptors in PBMC identified three subsets of cells with altered frequencies in SLE compared to healthy and autoimmune controls. We investigated the potential of each of these populations, taken individually or in combination, to distinguish SLE patients from healthy individuals and patients with other autoimmune diseases. The populations of interest were present in healthy and autoimmune disease individuals. Therefore, we used their frequencies and evaluated them as continuous variables. To compare SLE and HC, the frequency of each cell subset was normalized (min-to-max normalization). Then, we determined which cell populations taken separately was the best marker to differentiate SLE patients from healthy controls. ROC curves showed that the measurement of the SLEcTFH population was the best individual marker (AUC = 0.92) to distinguish SLE from HC ( Figure 6A ). Then, we combined the normalized values of the different populations to determine which combination best discriminates SLE from HC. We observed that the measurements of SLESMB together with SLEcTFH increases the performance from SLESMB AUC = 0.83, SLEcTFH AUC = 0.92 to AUC = 0.94 ( Figure 6A ).

Secondly, we examined which cell population (normalized frequencies) best distinguishes SLE from other autoimmune diseases using subjects included in the cohort 2. We observed that the single measurement of SLEB1 and SLESMB better discriminates SLE from autoimmune controls compared to SLEcTFH (SLEB1/SLESMB AUC = 0.81 vs. SLEcTFH AUC = 0.72, Figure 6B and Supplementary Figure 8 ). Furthermore, the combined measurements of SLEB1 and SLEcTFH taken together was the best to differentiate SLE from autoimmune diseases (AUC = 0.847, Figure 6B and Supplementary Figure 8 ). We then calculated the ideal cut-off to distinguish SLE from autoimmune diseases controls in cohort 2. Using the Youden index, we determined that individuals with a score greater than 0.282 (for both the SLEB1-SLEcTFH and SLESMB-SLEcTFH combinations) can be diagnosed as SLE. Furthermore, we evaluated whether the measurements of these cell populations were associated with specific clinical characteristics of SLE (cohort 1). Our data suggest that the measurement of SLEcTFH and the combined measurement of SLESMB-SLEcTFH discriminate patients with arthritis ( Supplementary Figure 9 ). Nevertheless, a larger sample size, presenting varying organ involvement, is necessary to confirm these associations. Overall, our data indicate that the combination of SLEB1 and SLEcTFH measurements or the combination of SLESMB and SLEcTFH, both correctly diagnosed 90% of SLE samples ( Figure 6C and Supplementary Table 1C ). These results show that the expression of SLAMF receptors by PBMC can represent a powerful diagnostic tool for SLE.